US 6298162 B1 Résumé A fast method with logarithmic complexity for image compression and expansion is provided, implementable on the Geometric Arithmetic Parallel Processor (GAPP) or any similar fine-grained, mesh-connected Single Instruction-Multiple Datastream (SIMD) array processor. Efficient decomposition of the original image into multiple subimages enables the method to exploit the parallelism of the SIMD computer for implementing powerful multiresolution pyramid and hierarchical image processing techniques on a massively parallel processor.
Revendications(10) 1. A method for implementing a compressive transformation of a first representation of a pattern of information made up of a plurality of picture elements into a second representation of the pattern of information, the method comprising the steps of:
first parallel aggregation of selected ones of said plurality of picture elements into first aggregates, wherein each of said first aggregates includes a copy of more than one of said picture elements; and
second parallel aggregation of said first aggregates into second aggregates, wherein each second aggregate includes a copy of more than one of said first aggregates.
2. A method for implementing an expansive transformation of a first representation of a pattern of information made up of a plurality of picture elements into a second representation of the pattern of information, the method comprising the steps of:
first parallel separation of selected ones of said plurality of picture elements into first aggregates, wherein each of said first aggregates includes copies of fewer of said plurality of picture elements than said first representation; and
second parallel separation of said first aggregates into second aggregates, wherein each second aggregate includes copies of fewer of said plurality of picture elements than each of said first aggregates.
3. A device for transforming a first representation of an information pattern made up of a plurality of picture elements into a second representation of the information pattern, the device comprising:
first means for parallel aggregation of selected ones of said plurality of picture elements into first aggregates, wherein each of said first aggregates includes a copy of more than one of said picture elements; and
second means for parallel aggregation of said first aggregates into second aggregates, wherein each second aggregate includes a copy of more than one of said first aggregates.
4. A device for implementing a compressive transformation of a first representation of a pattern of information made up of a plurality of picture elements into a second representation of the pattern of information, the device comprising:
first means for parallel aggregation of selected ones of said plurality of picture elements into first aggregates, wherein each of said first aggregates includes a copy of more than one of said picture elements; and
second means for parallel aggregation of said first aggregates into second aggregates, wherein each second aggregate includes a copy of more than one of said first aggregates.
5. A device for implementing an expansive transformation of a first representation of a pattern of information made up of a plurality of picture elements into a second representation of the pattern of information, the device comprising:
first means for parallel separation of a set of said plurality of picture elements of said first representation into first subsets, wherein each of said first subsets includes copies of fewer of said picture elements than said set; and
second means for parallel separation of said first subsets into second subsets, wherein each of said second subsets includes copies of fewer picture elements than said first subset from which said second subset was obtained.
6. A method of scaling a digital representation of an image, said image being divided into M rows and N columns of P picture elements, M and N being positive integers and P being the product of M and N, said digital representation comprising P digital values each respectively indicative of a characteristic of an associated one of said P picture elements, into a scaled digital representation of said image, said scaled digital representation comprising Q digital values, Q being the product of P and a scaling factor L, said method comprising the steps of:
loading said digital representation into a processor comprising an array of at least M rows and N columns of processing elements so that each of said processing elements stores an associated loaded one of said P digital values as a stored value;
controlling each row of at least two pairs of rows of processing elements, paired rows being separated by S rows, S being related to L, to operate in parallel to cause each processing element in said each row to replace its associated loaded value as its stored value with the associated loaded digital value from a processing element in the same column at a preselected number of rows away, at least one row of each pair being one of said M rows; and
controlling each column of at least two pairs of columns of processing elements, paired columns being separated by S columns, to operate in parallel to cause each processing element in said each column to replace its associated loaded value as its stored value with the associated loaded digital value from a processing element in the same row at a preselected number of columns away, at least one column of each pair being one of said N columns.
7. The method of claim
6, wherein said scaling is a compression of said digital representation of an image, said image being divided into M rows and N columns of P picture elements, M and N being positive integers and P being the product of M and N, said digital representation comprising P digital values each respectively indicative of a characteristic of an associated one of said P picture elements, into a scaled digital representation of said image, said scaled digital representation comprising Q digital values, Q being the product of P and a scaling factor L, said method comprising the steps of:loading said digital representation into a processor comprising an array of at least M rows and N columns of processing elements so that each of said processing elements stores an associated loaded one of said P digital values as a stored value;
controlling each row of at least two pairs of rows of processing elements, paired rows being separated by S rows, S being related to L, to operate in parallel to cause each processing element in said each row to replace its associated loaded value as its stored value with the associated loaded digital value from a processing element in the same column at a preselected number of rows away, at least one row of each pair being one of said M rows; and
controlling each column of at least two pairs of columns of processing elements, paired columns being separated by S columns, to operate in parallel to cause each processing element in said each column to replace its associated loaded value as its stored value with the associated loaded digital value from a processor in the same row at a preselected number of columns away, at least one column of each pair being one of said N columns.
8. A device for transforming a first representation of an information pattern made up of a plurality of picture elements into a second representation of the information pattern, the device comprising:
first means for parallel aggregation of selected ones of said plurality of picture elements into first aggregates, wherein each of said first aggregates includes a copy of more than one of said plurality of picture element; and
second means for parallel aggregation of said first aggregates into second aggregates, wherein each second aggregate includes a copy of more than one of said first aggregates;
wherein said first and second means for parallel aggregation further comprise a parallel data processor having a plurality of substantially identical cells for processing digital data signals, said parallel data processor further comprising:
a controller for generating control signals in response to program instructions; and
a plurality of substantially identical interconnected cells each including a full adder having two data input terminals, a carry input terminal, a sum output terminal, and a carry output terminal, and a plurality of memories connected to said full adder and said controller, the memories each being individually controllable to supply selected ones of a plurality-of predetermined data signals to the input terminals of said full adder in response to control signals from the controller, the memories being connected to the full adder and the controller such that both logical and arithmetic operations are performed by the full adder.
9. A device for transforming a first representation of an information pattern made up of a plurality of picture elements into a second representation of the information pattern, the device comprising:
first means for parallel aggregation of selected ones of said plurality of picture elements of said first representation into first aggregates, wherein each of said first aggregates includes a copy of more than one of said plurality of picture elements; and
second means for parallel aggregation of said first aggregates into second aggregates, wherein each second aggregate includes a copy of more than one of said first aggregates;
wherein said first and second means for parallel aggregation further comprise a parallel data processor having a plurality of substantially identical cells for processing digital data signals, said parallel data processor further comprising:
a controller for generating control signals; and
a plurality of identical cells, each of said identical cells being connected to at least one neighboring cell and responsive to said controller, and adapted to process data from said at least one neighboring cell in accordance with said control signals, said cell including a memory segmented into at least two blocks to be able to perform more than one read/write operation per clock cycle at the direction of said controller.
10. A device for transforming a first representation of an information pattern made up of a plurality of picture elements into a second representation of the information pattern, the device comprising:
first means for parallel aggregation of selected ones of said plurality of picture elements of said first representation into first aggregates, wherein each of said first aggregates includes a copy of more than one of said plurality of picture elements; and
second means for parallel aggregation of said first aggregates into second aggregates, wherein each second aggregate includes a copy of more than one of said first aggregates;
wherein said first and second means for parallel aggregation further comprise a parallel data processor having a plurality of substantially identical cells for processing digital data signals, said parallel data processor further comprising:
a controller for generating control signals; and
a plurality of identical cells, each of said identical cells being connected to at least one neighboring cell and responsive to said controller, and adapted to process data from said at least one neighboring cell in accordance with said control signals, said cell including a memory having at least two ports to be able to perform more than one read/write operation per clock cycle at the direction of said controller.
Description This application is a continuation, of application Ser. No. 07/995,696, filed Dec. 23. 1992, now abandoned. The present invention pertains to implementation of a method for fast image compression and expansion in a Single Instruction-Multiple Datastream (SIMD) array processor, and provides real-time spatial resolution image compression and expansion processing. Spatial resolution compression and expansion are necessary in multiresolution image processing. Existing methods for image compression and expansion require too many instructions, and may not be suitable for certain types of real-time applications. For the purpose of this discussion, the terms compression and expansion have the following meanings. Compression means sub-sampling or reducing the spatial resolution of the given image by a factor K, where K is some small positive integer. It is equivalent to low-pass filtering. Expansion means magnifying or increasing the size of the given image by a factor K. For convenience, the methods will be explained using an example with K=2. Let I(m,n) be the original image containing 2M×2N pixels, where m=1, 2, . . . , 2M; n=1, 2, . . . 2N; M and N are positive integers. The compression of I(m,n) by a factor of 2 in both dimensions of the image yields
where C(j,k) is the compressed image; j=1, 2, . . . , M; and k=1, 2, . . . , N. Some applications may require local averaging to be performed on I(m,n) prior to the compression. There are varying methods for achieving expansion or compression. One type of method is a sequential method; another is a parallel method such as can be implemented on a parallel processor. The complexity of a sequential method may be expressed as O(M×N), on the order of magnitude of M×N or proportional to M×N. The sequential method involves the copying of M×N selected pixels, one pixel at a time, from I(2j,2k) to C(j,k). Using a parallel method, however, image compression is performed in two stages; first, with respect to rows, and then with respect to columns. To compress the rows, image I(2j,2k) is shifted until row 2j of image I(2j,2k) aligns with row j of compressed image C(j,k) before all pixels in row 2j are copied simultaneously into compressed image C(j,k), as shown in FIGS. General mathematical properties of image compression and other digital image processing techniques are presented, for example, in The parallel method is an example of certain data processing tasks that require that substantially identical logical or arithmetical operations be performed on large amounts of data. One approach to carrying out such tasks which is drawing increasing attention is massively parallel processing. In parallel processing, each element or cell of an array processor made up of such cells processes its own bit of data at the same time as all other cells of the array processor perform the same process on their own bit of data. Such machines are referred to by several names, including Single Instruction-Multiple Datastream (SIMD), and massively parallel array processors. A common arrangement for such a machine is as a rectangular array of cells, with each interior cell connected to its four nearest neighboring cells and each edge cell connected to a data input/output device. Each cell is connected as well to a master controller which coordinates the movement of data through the array by providing appropriate instructions to the processing elements. When used in high resolution image processing, such processors handle the image pixels as a data matrix which can be loaded into and processed quickly and efficiently by the processor array. Although based upon the same generic concept of an array of cells all performing the same function in unison, parallel processors vary in details of cell interconnection and design. For example, U.S. Pat. No. 4,215,401 to Holsztynski et al. discloses a cell which includes a random access memory (RAM), a single bit accumulator, and a simple logical gate. The disclosed cell is extremely simple and, hence, inexpensive and easily fabricated. A negative consequence of this simplicity, however, is that some complex computational operations are quite cumbersome so that it may require many basic operations (instructions) to perform a complex task. U.S. Pat. No. 4,739,474, to Holsztynski et al., represents a higher level of cell complexity, in which the logic gate is replaced by a full adder capable of performing both arithmetic and logical functions. Pressing the full adder into dual service creates an efficiency which more than offsets the added complexity and cost incurred by including a full adder in each cell. Parallel processors may also vary in the manner of cell interconnection. As mentioned above, cells are typically connected to their nearest physical neighbors. All cells except those at the edge of the entire array are connected to four neighbors. Significant benefits may, however, flow from providing for alternate paths of interconnection and, specifically, in providing programmable, flexible interconnection between cells. A fast method with logarithmic complexity for image compression and expansion is provided, implementable on a fine-grained, mesh-connected Single Instruction-Multiple Datastream (SIMD) array processor such as Martin Marietta's Geometric Arithmetic Parallel Processor (GAPP). Efficient decomposition of the original image into multiple subimages exploits the parallelism of the SIMD computer for implementing powerful multiresolution pyramid and hierarchical image processing techniques. While the complexity (execution time) of known parallel methods is O(M+N), the complexity of the new fast parallel method of the present invention may be expressed as O(log The fast image compression and expansion implementations of the present invention are suitable for a parallel data processor comprised of an array of identical cells concurrently performing identical operations under the direction of a central controller, and incorporating one or more special cell architectures such as a segmented memory, conditional logic for preliminary processing, circuitry for indicating when the cell is active, and programmable cell interconnection including cell bypass and alternate connection of edge cells. The fast image compression and expansion implementations of the present invention are also suitable for a single instruction-multiple datastream (SIMD) systolic array processor having provision for local address generation, direct access to external devices, and programmable cell interconnectivity for providing great versatility while at the same time retaining the advantages of the SIMD architecture. Applications include use of the fast image compression and expansion methods of the present invention for compressing and/or expanding optical images in both one and two dimensions. Results to date demonstrate compression superior (several times faster) to previously attempted compression methods. The improvement in compression (or expansion) speed will increase with the size of the original image. In one aspect, the present invention provides a device and method for transforming a first representation of an information pattern, where the first representation is characterized by a first size, into a second representation of the information pattern, where the second representation is characterized by a second size different from the first size. The exemplary device and method include a first process for parallel aggregation of irreducible quantized elements (for example, pixel values) of the first representation into first aggregates, where each first aggregate includes more than one irreducible quantized element. The exemplary device and method further include a second process for parallel aggregation of the first aggregates into second aggregates, where each second aggregate includes more than one first aggregate. In another aspect, the present invention provides a method for scaling a digital representation of an image, where the image is divided into M rows and N columns of P picture elements, M and N are positive integers, and P is the product of M and N. The digital representation includes P digital values each respectively indicative of a characteristic of an associated one of the P picture elements. The scaled digital representation of the image includes Q digital values, where Q is the product of P and a scaling factor L. The method includes loading the digital representation into a processor including an array of at least M rows and N columns of processing elements so that each of the processing elements stores an associated one of the P digital values. The method further includes controlling at least two pairs of rows of processing elements, where paired rows are separated by S rows, where S is related to L, to operate in parallel, at least one row of each pair being one of the M rows, and controlling at least two pairs of columns of processing elements, where paired columns are separated by S columns, to operate in parallel, at least one column of each pair being one of the N columns. In yet another aspect, the present invention provides a device for transforming a first representation of an information pattern, where the first representation is characterized by a first size, into a second representation of the information pattern, where the second representation is characterized by a second size different from the first size. The device includes a first mechanism for parallel aggregation of irreducible quantized elements of the first representation into first aggregates, where each first aggregate includes more than one irreducible quantized element. The device further includes a second mechanism for parallel aggregation of the first aggregates into second aggregates, where each second aggregate includes more than one first aggregate. The first and second mechanisms for parallel aggregation further include a parallel data processor having a plurality of substantially identical cells for processing digital data signals. The parallel data processor further includes a controller for generating control signals in response to program instructions, and a plurality of substantially identical interconnected cells each including a full adder having two data input terminals, a carry input terminal, a sum output terminal, and a carry output terminal, and a plurality of memories connected to the full adder and the controller. The memories are each individually controllable to supply selected ones of a plurality of predetermined data signals to the input terminals of the full adder in response to control signals from the controller. The memories are connected to the full adder and the controller in such a way that both logical and arithmetic operations are performed by the full adder. Alternatively, an exemplary parallel data processor includes a controller for generating control signals, and a plurality of identical cells. Each of the identical cells is connected to at least one neighboring cell and is responsive to the controller, and is adapted to process data from at least one neighboring cell in accordance with the control signals. Each cell includes a memory segmented into at least two blocks to be able to perform more than one read/write operation per clock cycle at the direction of the controller. Alternatively, another exemplary parallel data processor includes a controller for generating control signals, and a plurality of identical cells. Each of the identical cells is connected to at least one neighboring cell and is responsive to the controller, and is adapted to process data from at least one neighboring cell in accordance with the control signals. Each cell includes a memory having at least two ports to be able to perform more than one read/write operation per clock cycle at the direction of the controller. The foregoing and other features of the present invention will become apparent to one of ordinary skill in the art to which the invention pertains from the following detailed description when read in conjunction with the drawings, in which: FIG. 1 is a block diagram of a preferred embodiment of a processor array useful for implementing the present invention; FIG. 2 is a schematic diagram of a second preferred embodiment of a processor array useful for implementing the present invention; FIGS. FIG. 4 is a block diagram of a preferred embodiment of a cell in FIG. 1 for a parallel bit serial data processor useful for implementing the present invention; FIG. 5 is a functional block diagram of another preferred embodiment of a cell for a parallel bit serial data processor useful for implementing the present invention; FIG. 6 is a diagram illustrating an operation of the local address generation function of the cell of FIG. 4; FIGS. FIG. 8 is a functional block diagram of interconnection of a cell with a coprocessor useful for implementing another preferred embodiment of the present invention; FIG. 9 is a schematic diagram showing a pattern of cell interconnection using cells useful for implementing a preferred embodiment of the present invention; FIG. 10 is a functional block diagram of a 4×4 array of cells; FIG. 11 is a key for showing I/O connections for a typical cell in the array of FIG. 10; FIGS. FIGS. FIGS. FIGS. The fast image compression and expansion processing of the present invention can be implemented most advantageously in a single instruction-multiple datastream (SIMD) machine such as the Geometric Arithmetic Parallel Processor (GAPP) described in commonly assigned U.S. Patent application Ser. No. 07/254,718. The GAPP is a fine-grained, mesh-connected computer, wherein each processor element is a 1-bit serial arithmetic logic unit (ALU) with 128 bits of random access memory (RAM). Before describing the fast image compression and/or expansion processing of the present invention, several embodiments of SIMD array processors are described that are useful for implementing the present invention. General details concerning the arrangement of parallel processors as an array of identical cells in the Geometric-Artihmetic Parallel Processor (GAPP) are available from aforementioned U.S. Pat. No. 4,739,474 to Holsztynski et al., hereinafter the “Holsztynski patent”, the specification of which is incorporated here by reference. More specific details concerning arrangements of parallel processors as arrays of identical cells that are particularly useful for implementing the present invention are disclosed in the following co-pending patent application and issued patents (all assigned to the present assignee): U.S. Ser. No. 07/254,718, filed Oct. 7, 1988, directed to a GAPP; U.S. Patent No. 5,179,714 to Robert B. Graybill, hereinafter the “Graybill patent”, directed to an Advanced Systolic Array Processor (ASAP); and U.S. Patent No. 5,173,947 to Chandé et al., hereinafter the “Chandé patent”, directed to a conformal image processor. The specifications of the Graybill and Chandé patents are incorporated here by reference. Features of these systems which make them especially useful include their north, south, east, and west connectivity (with wraparound capability), and the capability for internal arithmetic and logic operations. Additionally, ASAP systems are endowed with extensive internal memory for each cell, and, like the later GAPP systems, the capability within each cell for local address generation. The present invention will therefore be described as it would be implemented in these systems. Advantages in the processing speed that fast image compression and expansion methods implemented on massively parallel SIMD array processors have over prior methods increase with the size of the image to be compressed or expanded. A GAPP with a higher density of processors per chip than other known processor chips is particularly suitable for implementing a large array for a large image size. It will be understood by one having ordinary skill in the art, however, that the invention could also be implemented in other systems. As shown in the Holsztynski patent and in FIG. 1 herein, an array The edge cells of the array FIG. 1 also shows two multiplexer networks intermeshed with the cell array. The first multiplexer network is made up of multiplexers M The second multiplexer network is made up of multiplexers M All of the multiplexers are responsive to the controller FIG. 2 is a more detailed rendering of an array When cell bypass is invoked, cell-to-cell communication is based on the value of a latched bypass control word which continuously drives the row and column multiplexer select lines. If it were desired to bypass the first column, for example, the westernmost I/O pins become logically connected to the west port of the cells in the second column. Row and column bypass can be separately controlled so that all possible virtual array sizes, such as 7×12, 4×10, 1×16, or even 1×1, can be obtained. The array shown in FIG. 2, however, is designed so that it is not possible to bypass the array The benefits of the cell bypass network are well exemplified in operations such as parallel data summation, also known as an array sum. An array sum is a summation of the same data type variable in each cell throughout the entire array. With programmable bypass, this summation may be achieved using the following method: (1) program the cell bypass network to full 8×16 configuration; (2) command each cell to shift its variable to its eastern neighbor; (3) command each cell to add the received variable to its resident value (which results in cells in columns 2, 4, 6, and 8 containing a non-overlapping set of sums); (4) program the cell bypass network to a 4×16 configuration logically interconnecting columns 2, 4, 6, and 8; (5) repeat steps (2) and (3) (which results in cells in columns 4 and 8 containing a non-overlapping set of sums); (6) program the cell bypass network to a 2×16 configuration logically interconnecting cells in column 4 to cells in column 8; (7) repeat step 5 (which results in cells in column 8 containing the sums for the entire array from west to east); and (8) repeat the method north to south to sum cells in column 8, halving the row configuration after each shift and add until the entire array sum resides in the cell at the southeastern corner of the array. Without the cell bypass network, many additional clock cycles would be required to shift the data through the non-participating cells to the ultimate destination when performing the partial array summations. The cell bypass network allows direct cell communication for a minimal overhead of 2 cycles for array reconfiguration. The other multiplexer network comprises bidirectional semiconductor switches or multiplexers M The north alternate connector may also be connected to a memory device to extend logically the memory capability of each cell and to allow data operands to be rolled in and rolled out as necessary. It also provides a port which can be connected to an I/O device or other processing cells in the array to allow for corner turn operations. FIG. 4 is a block diagram of exemplary internal architecture of a cell As mentioned above, memory Only one segment may be written into during any given clock cycle. Which RAM segment is written into depends on control lines C The memory The segmented nature of an exemplary embodiment of the memory It has been mentioned above that the cell The SR/C It is also possible, however, to have the ALU In a more advanced utilization, the offset value or effective address may be calculated from the data set. This may prove useful, as described below, in the implementation of floating point arithmetic. This powerful but logically simple innovation thus allows independent addressing of the memory As an example of the usefulness of the local address generation capability created by the incorporation of SR/C The local addressing capability also makes it possible to implement true floating point arithmetic. An obstacle to the implementation of floating point arithmetic on prior SIMD machines is that the single instruction architecture is not amenable to operands being individually normalized prior to computation. The local address generation capability enables the exponent of a floating point number to be used as an index for a scratch memory location, thereby normalizing the operands prior to computation. The cell The CLU The CLU The sum, carry, and borrow output from the CLU The TD register One especially useful function of the communication line established through the TD register The TD registers There are many possible examples showing the usefulness of the TD register The technique may be extended beyond floating point multiplication to any of the more complex, time consuming arithmetic operations such as square root and divide. This technique may be extended even further to the point of having multiple different computational memory resources sharing the same TD port. Basic methods such as singular value decomposition (SVD) could be implemented more easily using the third dimension connections for additional external cell storage. Fast Fourier transform (FFT) signal processing could also be enhanced. Due to the symmetry of the FFT, the complex conjugate computation is really only a sign change and is folded back. Thus, only four horizontal rows of cells are required. The interstage shuffling of data is passed horizontally by nearest neighbor connections and diagonally by TD ports The AF register FIG. 10 shows a 4×4 array with cells having inputs as shown in FIG. The foregoing describes particular array processors which are especially well adapted to carry out the method of, and serve as part of the apparatus of, the present invention. The principles of the invention will now be described in connection with the arrays. The above-described SIMD devices are particularly advantageous for implementing the multiresolution image processing of the present invention which is described in detail below. The method according to the present invention more fully exploits the parallelism of the SIMD machine. In the conventional method described above, effectively only one row (or column) is copied at a time, as shown in FIGS. For example, for scaling factor K=2, to obtain a compressed image containing rows 2, 4, 6, 8, . . . , 32 of the original image as shown in FIG. Although the new, fast parallel method implemented in the present invention requires about the same number of shift operations as the known parallel method, the number of copying operations required by the present invention is substantially reduced, in part because the parallel processing afforded by the parallel processing array is more fully exploited and the copying operations are carried out substantially in parallel. The total number of shift operations is about:
for a 2M×2N-pixel image, where k is the number of bits representing each pixel. This is also about the same as the number of copying operations required by the known parallel method. By way of contrast, the number of copying operations required for the new, fast parallel method implemented in the present invention is only about:
for a 2M×2N-pixel image, where k is the number of bits in each pixel, and where M and N are both greater than one. The complexity of the new, fast parallel method is O(log Depending on the applications, some noise reduction operator such as a mean or median filter can be applied to the original image before the compression, as described, for example, in the above-mentioned Gonzalez and Woods reference. In general, the scaling factor K can be any proper positive integer, and the rows and columns do not have to share the same K. A fast method for image magnification or expansion can be implemented on the GAPP array by performing the operations in a reverse order relative to the compression method. During the expansion, each row and column will be replicated a number of times determined by the appropriate scaling factor(s). In other words, the size of each pixel will be magnified. Depending on the applications, pixel-interpolation rather than direct replication may be used in the expansion. For example, an interpolation or extrapolation can be used to fill in blank pixels at the last step in the expansion from FIG. The method and apparatus of the present invention are especially useful in several areas. An area of particular usefulness is in real-time image processing. An exemplary application in this area will now be discussed. There is a need to provide real-time image processing in both the spatial and frequency domains based on fast image compression and/or expansion for image filtering, image analysis, better pattern recognition, image display, image registration, object tracking, bandwidth compression, superresolution, and multiresolution image processing. Referring again to FIG. 1, these capabilities may be realized if the data input/output device This is only one application. It will be apparent to one having ordinary skill in the art the invention will prove useful in additional applications as well. The present invention has been described above in terms of specific embodiments. It will be readily appreciated by one of ordinary skill in the art, however, that the invention is not limited to these embodiments, and that, in fact, the principles of the invention may be embodied and practiced in devices and methods other than those specifically described above. For example, it has already been mentioned that other geometries for the arrays may be useful. Furthermore, while the foregoing describes the processing of images using the present invention, one of ordinary skill in the art will recognize the usefulness of the present invention in the more general processing of any sensually apprehended patterns of information and intelligence. Therefore, the invention should not be regarded as being limited to these specific embodiments, but instead should be regarded as being fully commensurate in scope with the following claims. Furthermore, any element in the following claims expressed as a means or step for performing a specified function without the recital of structure, material, or acts in support thereof shall be construed to cover the corresponding structure, material, or acts described herein and equivalents thereof. Citations de brevets
Citations hors brevets
Référencé par
Classifications
Événements juridiques
Faire pivoter |